How to use the trained model to predict ？ about pykeen HOT 4 CLOSED

Hi,
you can load a trained model as follows (https://pytorch.org/tutorials/beginner/saving_loading_models.html):

import torch
path = 'path/to/trained_model.pkl'
model = torch.load(path)
model.eval()

Here we provide a tutorial of how to introspect a trained model and how to make predictions.

We provide further functions to make predictions, e.g., predict_scores(), predict_scores_all_tails(), and predict_scores_all_heads() which are defined in the base model class. However, these functions require that the entities and relations are already mapped to their unique ids (https://pykeen.readthedocs.io/en/latest/tutorial/performance.html#entity-and-relation-ids).
Entities and relations in triples can be mapped to their unique ids as follows:

import numpy as np
# Here, we assume that we trained the model on the 'Nations' dataset
triples = np.array([['brazil','conferences','china']])
mapped_triples = model.triples_factory.map_triples_to_id(triples=triples)

You then can, for instance, use the function predict_scores() to make predictions:

model.predict_scores(triples=mapped_triples)

If you want to map entities/relations individually, you can make use of the internal mappings model.triples_factory.entity_to_id and model.triples_factory.relation_to_id.

from pykeen.

A further example of how to make predictions can be found here.

from pykeen.

A further example of how to make predictions can be found here.

I would definitely suggest using the models from the tutorial from Mehdi's second post. If you're very comfortable with PyTorch, you might consider going into the guts he described in his first post.

from pykeen.

Thank you so much, great work ：）

from pykeen.